Evaluation of Reference Dilution Test Methods for Antimicrobial Susceptibility Testing of Pseudomonas aeruginosa Strains Isolated from Patients with Cystic Fibrosis

Abstract

The development of multidrug-resistant Pseudomonas aeruginosa in patients with cystic fibrosis (CF) is most likely a consequence of increasing life expectancy and more prolonged exposure to antibiotics. The optimal method for antibiotic susceptibility testing of CF strains, particularly mucoid P. aeruginosa strains, is unknown. Antimicrobial susceptibilities of 48 CF strains (25 mucoid) and 50 non-CF strains to 12 anti-Pseudomonas agents were tested by both agar dilution and commercially custom-prepared broth microdilution plates (PML Microbiologicals, Portland, Oreg.) in three laboratories simultaneously to determine if broth microdilution could substitute for agar dilution as the reference method in subsequent studies. Comparison of MICs generated by agar dilution and broth microdilution demonstrated correlation coefficients (r) exceeding 0.85 for all agents tested; correlation was excellent for aminoglycosides (r ≥ 0.92) and very good for β-lactam agents including agents paired with a β-lactamase inhibitor (r ≥ 0.87) and for ciprofloxacin (r = 0.86). Correlation was not improved by 48-h readings, but correlation between 24- and 48-h readings ranged between 0.91 and 0.98 for both methods. Interlaboratory variations were minimal, as the percentage of acceptable variations was 94% for both methods, and serious discords were infrequent (<2% of comparisons). However, CF strains were more likely to have serious discords than were non-CF strains (P < 0.0001), although mucoid strains were not more likely to have serious discords than were nonmucoid strains. In this study, MICs determined by custom-prepared broth microdilution compared favorably with MICs determined by agar dilution. Thus, this broth microdilution assay can serve as a reference method and facilitate future studies to determine the optimal method for antibiotic susceptibility testing of CF strains.

The life expectancy of patients with cystic fibrosis (CF) has dramatically increased over the past two decades in parallel with the development of antibiotics with activity against Pseudomonas aeruginosa (4, 11). Due to the frequent, prolonged antibiotic courses used to treat the pulmonary exacerbations which are the hallmark of CF, antibiotic-resistant P. aeruginosa strains develop. Despite a working definition for P. aeruginosa strains with resistance to multiple antibiotics (13), the prevalence of such strains isolated from CF patients has been examined in only one study (2). Multiple-antibiotic-resistant strains have obvious therapeutic implications for CF patients, but the optimal in vitro method for testing the antibiotic susceptibility of P. aeruginosa strains, especially mucoid strains, is unclear. In addition, individual patients may be referred to other CF centers and lung transplant programs that employ different antimicrobial susceptibility testing methods. Discrepancies in susceptibility patterns can confuse or potentially harm the management of the patient.

The objective of this study was to determine if commercially prepared custom broth microdilution (PML Microbiologicals, Portland, Oreg.) compared favorably with agar dilution and could serve as a more convenient reference method for subsequent studies.

MATERIALS AND METHODS

P. aeruginosa strains.

Forty-eight strains of P. aeruginosa from individual CF patients residing in 37 states in the United States (23 nonmucoid and 25 mucoid strains) were studied. These strains were derived from the collection at the CF Referral Center at Columbia University. Fifty non-CF strains, one from each of 41 states and nine from the United Kingdom, from the research clinical microbiology laboratory at the University of Iowa Hospital and Clinics were also tested. Strains were grown on blood agar plates and MacConkey agar plates to assess purity and on Mueller-Hinton agar plates (Becton Dickinson, Cockeysville, Md.) to assess pigment production.

Confirmation of identification as P. aeruginosa.

All strains were probed for the exotoxin A gene to confirm their identification as P. aeruginosa (14). The EcoRI-HindIII fragment of plasmid pRGI containing the exoA gene from P. aeruginosa (kindly provided by S. Lory, University of Washington, Seattle) was isolated and labeled with digoxigenin according to the manufacturer’s instructions (Genius kit, Boehringer Mannheim, Indianapolis, Ind.). The strains were grown overnight on blood agar plates. A single colony was replica plated to a nylon transfer membrane (MSI, Westboro, Mass.) on the surface of a second blood agar plate, and colony hybridization was performed (15). Detection of the digoxigenin-labeled probe by chemiluminescence was performed according to the manufacturer’s recommendations (Boehringer Mannheim). The strains that did not have the gene for exotoxin A were further evaluated by the API 20 NE system (bioMerieux Vitek, Inc.).

Antibiotics studied.

Susceptibility to the following 12 antibiotics was tested: amikacin (2 to 256 μg/ml), gentamicin (0.5 to 64 μg/ml), tobramycin (0.5 to 64 μg/ml), ciprofloxacin (0.06 to 8 μg/ml), piperacillin (2 to 256 μg/ml), ticarcillin (2 to 256 μg/ml), piperacillin-tazobactam (2/4 to 256/4 μg/ml, respectively), ticarcillin-clavulanate (2/2 to 256/2 μg/ml, respectively), ceftazidime (0.5 to 64 μg/ml), aztreonam (0.5 to 64 μg/ml), imipenem (0.25 to 16 μg/ml), and meropenem (0.25 to 16 μg/ml). Antibiotics (amikacin, gentamicin, tobramycin, ticarcillin, and ceftazidime) were obtained from Sigma Chemical Company (St. Louis, Mo.) with the exception of ciprofloxacin (Bayer Diagnostic Pharmaceuticals, Kankakee, Ill.), piperacillin-tazobactam (Lederle Laboratories, Wayne, N.J.), ticarcillin-clavulanate (SmithKline Beecham, Knoxville, Tenn.), aztreonam (Bristol-Myers Squibb, Princeton, N.J.), imipenem (Merck & Co., Inc., West Point, Pa.), and meropenem (Zeneca Pharmaceuticals, Wilmington, Del.), which were obtained from their respective manufacturers.

Agar dilution.

Cation-adjusted Mueller-Hinton agar plates containing serial twofold dilutions of antibiotics were prepared in each of three participating laboratories (Columbia University, Children’s Hospital and Regional Medical Center, and University of Iowa) with the same lots of antibiotics. Plates were inoculated with a Steers replicator (Clathra Systems MCT Medical, St. Paul, Minn.) according to National Committee for Clinical Laboratory Standards standard M7-A3 1993 (10), incubated at 35°C, and read at 18 to 24 and 48 h to accommodate slower-growing strains. Susceptibility interpretations were made according to National Committee for Clinical Laboratory Standards standards (10).

Broth microdilution plates.

Broth microdilution plates were custom prepared for this study by PML Microbiologicals and inoculated according to the recommendations of the manufacturer. Briefly, pure colonies grown on blood agar were used to inoculate 3 ml of sterile water to a 0.5 McFarland standard. One milliliter of this suspension was added to 29 ml of sterile water to provide the inoculum for the microdilution trays, which were inoculated with a multipoint inoculator supplied by the manufacturer. Plates were incubated at 35°C and read at 18 to 24 and 48 h to accommodate slower-growing strains. The MIC was considered the lowest concentration of an antimicrobial agent that completely inhibited growth in the microtiter well as detected by eye with a viewing device that facilitated reading. Susceptibility interpretations were made according to National Committee for Clinical Laboratory Standards standards (10).

Data analysis.

All data was entered into Microsoft Excel 97 SR-1 (Microsoft, Seattle, Wash.), and correlation coefficients of the MICs for each antibiotic by each method for readings at both 24 and 48 h were generated. An ideal statistical result demonstrating perfect correlation is a 1.00 correlation coefficient and a 1.00 slope. Variations among the three participating laboratories and discordant results (differences ≥ 2 log₂ dilution steps) among susceptibility testing methods and laboratories were analyzed. In addition, the frequencies of serious discords in the categorization of susceptibility, i.e., susceptible versus resistant, were determined for each laboratory, for CF strains compared with non-CF strains, and for nonmucoid strains compared with mucoid strains with a chi-square analysis.

RESULTS

Strain identification.

Six strains did not have the gene for exotoxin A (18). Further identification with the API 20 NE system confirmed that these six strains were P. aeruginosa.

Comparison of agar dilution and broth microdilution methods.

Comparison of the MICs generated by agar dilution and broth microdilution methods demonstrated correlation coefficients and regression slopes ≥0.85 at 24 h for all the antimicrobial agents tested. The pooled results for all three laboratories at 24 and 48 h are shown in Table 1. There was excellent correlation between methods for aminoglycosides (r ≥ 0.92) and very good correlation for β-lactam agents, including testing with β-lactamase inhibitors (r ≥ 0.87), and for ciprofloxacin (r = 0.86). Correlation of readings made at 48 h did not improve the correlation between susceptibility methods, and correlations were slightly lower than those made at 24 h for half of the agents studied.

TABLE 1.

Comparison of agar dilution and broth microdilution method MICs for the three participating laboratories interpreted at 24 and 48 h^a

Antimicrobial agent	Value at 24 h/value at 48 h
	Correlation coefficient (r)	Regression slope
Amikacin	0.92/0.92	0.98/0.95
Gentamicin	0.96/0.96	1.00/1.00
Tobramycin	0.94/0.93	1.02/1.00
Ciprofloxacin	0.86/0.86	0.86/0.87
Piperacillin	0.88/0.83	0.89/0.91
Ticarcillin	0.89/0.88	0.92/0.86
Piperacillin-tazobactam	0.88/0.92	0.93/0.98
Ticarcillin-clavulanate	0.87/0.88	0.96/0.98
Ceftazidime	0.91/0.93	0.93/0.90
Aztreonam	0.88/0.87	0.87/0.86
Imipenem	0.88/0.85	0.85/0.76
Meropenem	0.88/0.87	0.87/0.87

^an = 6,999 comparisons.  

Comparison between 24- and 48-h readings.

Correlation between the 24- and 48-h readings was excellent for both test methods and ranged between 0.91 and 0.98 (Table 2). Correlations generated by the agar dilution method were generally the same as or only 0.02 better (mean r = 0.95) than correlations generated by the broth microdilution method (mean r = 0.94) with the exception of that for aztreonam. For this agent, the correlation between the two time interval readings was 0.97 for agar dilution and 0.91 for broth microdilution.

TABLE 2.

Comparison of 24- and 48-h results by method and by antimicrobial agent^a

Antimicrobial agent	Agar dilution		Broth microdilution
	r	Slope	r	Slope
Amikacin	0.96	1.00	0.95	0.96
Gentamicin	0.98	0.97	0.96	0.96
Tobramycin	0.97	0.98	0.95	0.95
Ciprofloxacin	0.93	0.95	0.94	0.96
Piperacillin	0.92	0.82	0.95	0.91
Ticarcillin	0.93	0.91	0.93	0.87
Piperacillin-tazobactam	0.93	0.88	0.93	0.89
Ticarcillin-clavulanate	0.93	0.85	0.93	0.88
Ceftazidime	0.97	0.96	0.95	0.90
Aztreonam	0.97	0.95	0.91	0.88
Imipenem	0.95	0.89	0.93	0.81
Meropenem	0.94	0.88	0.94	0.87

^an = 7,010. r = correlation coefficient. 

Interlaboratory variations.

The number of MIC determinations generated by the three participating laboratories that resulted in variations, i.e., those MICs that were more than 1 log₂ dilution from the geometric mean, was determined (Table 3). Results that yielded variations were very similar for both methods and for readings generated at either 24 or 48 h. The mean percentage of comparisons with acceptable variations, i.e., within 1 log₂ dilution from the geometric mean, was identical for agar and broth, 94.1 and 94.2%, respectively. The most variation occurred with piperacillin with and without the β-lactamase inhibitor tazobactam. However, variations beyond acceptable limits occurred infrequently and were distributed equally between 24- and 48-h readings, although variations generated by laboratory A for the broth microdilution method were higher (Table 4).

TABLE 3.

Interlaboratory variations among three participating laboratories by two methods with readings of MICs at 24 and 48 h^a

Antimicrobial agent	Agar dilutionb		Broth microdilution		% Acceptable variationc
	24 h	48 h	24 h	48 h	Agar dilution	Broth microdilution
Amikacin	9	9	6	8	96.9	97.6
Gentamicin	10	13	14	12	96.1	95.6
Tobramycin	14	10	13	13	95.9	95.6
Ciprofloxacin	14	15	18	13	95.1	94.7
Piperacillin	22	25	20	22	90.8	92.9
Ticarcillin	17	17	18	15	94.2	94.4
Piperacillin-tazobactam	27	27	30	20	90.8	91.5
Ticarcillin-clavulanate	16	13	17	13	95.1	95.0
Ceftazidime	20	17	24	16	93.7	93.2
Aztreonam	21	16	26	18	93.7	92.5
Imipenem	20	20	18	14	93.2	94.6
Meropenem	15	19	18	21	94.2	93.4
Total no. of variations	205	201	222	185	94.1 (mean)	94.2 (mean)

^aVariations are results more than 1 log₂ dilution from the geometric mean MIC generated by the three participating laboratories. 

^bThe numbers of variations are provided for agar dilution and broth microdilution readings at 24 and 48 h which reflect 294 observations (98 isolates studied by three laboratories) for each agent for each method at each time point. 

^cResults in the three laboratories within 1 log₂ dilution from the geometric mean MIC. 

TABLE 4.

Occurrence of variations of MICs beyond acceptable limits (±1 log₂ dilution) of geometric mean MIC^a

Laboratory	Agar dilution		Broth microdilution
	24 h	48 h	24 h	48 h
A	79	77	120	101
B	58	51	50	34
C	68	73	52	50
Subtotal	205	201	222	185

^an = 14,112 comparisons. The number of variations beyond acceptable limits is provided. Totals are 406 for agar dilution and 407 for broth microdilution. 

Variations resulting in serious discords.

The number of serious discords, i.e., variations in MICs that alter the categorization of susceptibility versus resistance, was calculated. As shown in Table 5, less than 2% of all results led to serious discords, although laboratory A had slightly more discordant results. Of note, CF strains, both mucoid and nonmucoid, were twice as likely to have serious discords as were non-CF strains (odds ratio, 1.89 [95% confidence intervals of 1.38 and 2.59]; P < 0.0001).

TABLE 5.

Variations resulting in serious discords in the categorization of susceptibility (susceptible versus resistant)^a

Antimicrobial agent	No. of serious discords by:
	Participating laboratory			Organism groupb
	A	B	C	Nonmucoid	Mucoid	Non-CF
Amikacin	1	1	3	3	2	0
Gentamicin	4	0	2	2	2	2
Tobramycin	3	1	1	1	3	1
Ciprofloxacin	6	6	4	10	3	3
Piperacillin	15	4	6	6	9	10
Ticarcillin	12	4	7	5	10	8
Piperacillin-tazobactam	11	13	6	9	12	9
Ticarcillin-clavulanate	9	3	5	5	7	5
Ceftazidime	5	3	3	0	5	6
Aztreonam	9	5	4	5	7	6
Imipenem	5	5	8	9	2	7
Meropenem	7	2	4	3	2	8
No. of observations	4,704	4,704	4,704	3,312	3,600	7,200
Total no. (%)	87 (1.8)	47 (1.0)	53 (1.1)	58 (1.8)	64 (1.8)	65 (0.9)

^aAnalysis includes the number of variations for agar dilutions and broth microdilutions read at 24 and 48 h. 

^bNonmucoid and mucoid strains are from CF patients. 

Intralaboratory reproducibility for agar dilution and broth microdilution methods.

Analysis of intralaboratory reproducibility for both methods was performed by laboratory A. The majority of MICs were the same or within 1 log₂ dilution in each of the triplicate assays. Only 2.4 and 2.2% of agar dilution and broth microdilution studies, respectively, yielded unacceptable variation (Table 6). More than half of the variations occurred with aminoglycosides.

TABLE 6.

Intralaboratory reproducibility for agar dilution and broth microdilution methods by laboratory A^a

Antimicrobial agent	Agar dilution				Broth microdilution
	0	±1	±2	>±2	0	±1	±2	>±2
Amikacin	67	18	0	5	72	13	2	3
Gentamicin	80	6	0	4	80	5	0	5
Tobramycin	74	10	0	6	76	12	1	1
Ciprofloxacin	81	9	0	0	76	14	0	0
Piperacillin	78	9	0	3	77	12	1	0
Ticarcillin	77	10	3	0	81	9	0	0
Piperacillin-tazobactam	78	12	0	0	77	12	1	0
Ticarcillin-clavulanate	75	15	0	0	79	10	1	0
Ceftazidime	74	16	0	0	82	8	0	0
Aztreonam	69	21	0	0	78	10	2	0
Imipenem	77	9	1	3	82	7	1	0
Meropenem	81	9	0	0	75	10	5	0
Total	911	144	4	21	935	122	14	9

^aThere were 2,160 comparisons. The number of variations is provided. 

DISCUSSION

It is critical to determine the optimal method for accurate antibiotic susceptibility testing for CF strains of P. aeruginosa. The clinical implications of antibiotic resistance are considerable and include antimicrobial treatment, infection control, and transplant eligibility (1, 6). This issue is confounded by the multiple morphotypes of P. aeruginosa present in CF sputum (2, 3, 8, 9). In 1994, the U.S. Cystic Fibrosis Foundation convened an expert advisory panel to develop guidelines for laboratories affiliated with Cystic Fibrosis Foundation-approved centers (12). Experts agreed that selective medium was imperative for correctly isolating all potential pathogens (5). While previous investigators have examined mixed morphotype susceptibility testing (3, 8, 9, 17), the panel agreed that such mixed morphotype testing could lead to inaccurate assessment of resistance (16). Finally, it was felt that antibiotic disk diffusion should be the preferred susceptibility testing method and that automated microtiter systems were inadequate to assess slowly growing and mucoid strains. However, it was agreed that data was needed in order to clearly endorse agar-based diffusion methods, to further examine the utility of the E-test (7), and to delineate the limitations of automated microtiter systems.

In this study, we have shown that custom-prepared broth microdilution plates could be used in place of the standard reference method of agar dilution. Variations, both inter- and intralaboratory, were comparable for both broth microdilution and agar dilution, and serious discords were minimal. Because of concern about slow-growing strains, we examined the correlation between 24 and 48 h and found it to be excellent. However, there is no clinical data to support 48-h susceptibility testing. CF strains were more likely to have serious discords than were non-CF strains, but somewhat surprisingly, there was not a difference in the number of serious discords found for mucoid strains and those found for nonmucoid strains. Due to the potential for slower growth and for the production of alginate by mucoid strains interfering with accurate preparation of the inocula and possibly obscuring endpoints, we had expected more discords attributable to mucoid strains. Perhaps the use of devoted laboratory technologists and manual MIC readings minimized the differences between mucoid and nonmucoid strains. However, this study suggests that CF strains are less reproducible than are non-CF strains. While further research should confirm this observation, it is possible that, in the future, clinical microbiology laboratories may have to use different protocols for antimicrobial susceptibility testing of P. aeruginosa strains isolated from CF patients.

Correlation coefficients were excellent for all aspects of comparison when agar dilution was compared to broth microdilution. It should be noted, however, that correlation coefficient calculations are favorably influenced by large denominators; with larger sample sizes, poor correlations have less impact on the final correlation coefficient. This study did employ a large number of comparisons, ranging from 2,000 to 7,200, which no doubt improved the comparability of these assays. However, such large numbers are unavoidable when working with numerous strains and multiple antibiotics.

Determining the suitability of broth microdilution to serve as a reference method will facilitate further studies of commercially available broth microdilution methods and agar-based diffusion methods. Such studies will also have important implications for future research involving CF patients. The development of a standardized, reference-quality, convenient method will facilitate studies of incidence and prevalence of antibiotic resistance, antibiotic treatment including the use of aerosolized agents, and the impact of other adjuvant therapies such as gene therapy on CF flora.